An AI agent in one day: a local prototype with no cloud and no developers

When management asks for an AI agent in a business process, the requirements usually contradict each other: data cannot go to the cloud, there is almost no budget, there are not enough developers, and the result is needed tomorrow. In practice, this is solvable. In one working day you can build a local prototype of a functional AI agent using open tools such as Ollama and n8n. You do not need a team of specialists, cloud subscriptions or a complex architecture.

Why local rather than cloud

Cloud LLM APIs such as OpenAI's are convenient, but the cost grows with every request, and the company's sensitive data goes to third parties. That is risky for organizations with confidentiality requirements. A local agent on the Ollama platform runs directly on the company's computer or server: data never leaves the perimeter, and you pay only for electricity.

Main advantages of the local approach:

• Data stays inside the company and does not go to the cloud
• No API bills, only one-time hardware costs
• You can work in a closed network without a constant internet connection
• Full independence from cloud services and their outages
• Cheaper to scale for large volumes of requests

Ollama and n8n: two tools for the build

Ollama packages large language models into a container. You download a ready-made model (Llama 2, Mistral, Deepseek, Phi and others), run it through Docker, and the model is available via a REST API. No Python, no CUDA setup, no fighting with dependencies. Within 15 minutes the model is ready to answer its first request.

n8n is a no-code automation platform. Think of it as a builder for workflows. You connect Ollama as a node in the visual editor, attach data sources (CRM, Slack, email, knowledge bases, files), create a chain of actions, and the agent starts working. You do not need to write code; everything happens in a drag-and-drop interface.

From zero to a working demo in a day

Here is a rough schedule for organizing this in one working day:

• 09:00 — install Ollama and download the chosen model (40–50 minutes)
• 09:50 — set up the local REST API, test the model with curl (30 minutes)
• 10:20 — install and launch n8n for the first time on the same machine (30 minutes)
• 10:50 — create the first workflow in n8n, connect Ollama as a node (1 hour)
• 11:50 — tune the prompt, test on simple examples (45 minutes)
• 12:35 — integrate a data source (for example, uploading documents or connecting to Slack) (1 hour)
• 13:35 — debug, fix errors, check edge cases (1 hour)
• 14:35 — demo the working prototype to management

This schedule is realistic if you do not obsess over perfection. The main thing is to show that the idea works.

When you need RAG: searching your own data

If the agent has to answer based on the company's internal information (reports, policies, technical docs, FAQs, sales histories), add RAG (retrieval-augmented generation). n8n can load documents, create their embeddings (vector representations) and, for each user question, search your knowledge base for relevant passages. The agent becomes significantly smarter because it works not just with what it learned in training, but with data specific to the company.

What this means in practice

Local AI agents are turning from experiments into working tools. A company of any size, from a startup to a corporation, can build a functional agent in a day that works with internal data and processes, without the risk of leaks to the cloud and without huge bills. This matters especially for the financial sector, government bodies and manufacturing, where data confidentiality is not a preference but a strict requirement of law and security policy.